Research map / Probabilistic and scientific modeling / Causal discovery

Identifiability of Direct Effects from Summary Causal Graphs

AAAI 2024 · Probabilistic and scientific modeling › Causal discovery · approach: Causal effect identification · 2 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Gives a complete graphical criterion for when a direct effect in a linear dynamic SCM is identifiable from a summary causal graph (cycles allowed), plus two finite adjustment sets for the identifiable cases.

Problem

Direct effects in linear dynamic structural causal models are usually identified from full-time causal graphs, but experts often only have summary causal graphs, which omit temporal information and allow cycles; identifiability from such graphs was not characterized for cycles longer than 2.

Approach

Derives a necessary-and-sufficient graphical condition for identifiability of direct effects from an SCG with arbitrary cycles, by reasoning over all compatible full-time causal graphs, and proposes two finite adjustment sets (one based on non-descendants and descendants of Y within the maximal lag, one restricted to ancestors of Y) that are valid whenever the effect is identifiable.

Key results

• Complete characterization: the direct effect alpha_{X_{t-gamma},Y_t} is non-identifiable from an SCG iff X is a parent of Y and either X is a descendant of Y with a cycle through X avoiding Y, or there is an active non-direct path whose interior lies in descendants of Y (with extra conditions when g
• Corollary: if the SCG contains a cycle of length greater than 2, some instantaneous direct effect is non-identifiable.
• Two finite adjustment sets are proposed; the second (restricted to ancestors of Y) is shown sound via a sketch proof.

Surprising findings

• Whenever an SCG has a cycle longer than 2, at least one instantaneous direct effect is non-identifiable, so cycles of length above 2 make identification fail in general.

Limitations

• Restricted to linear dynamic SCMs with no hidden confounding and stationarity.
• Assumes the full-time causal graph is acyclic.
• Does not characterize all possible valid adjustment sets; only two sound sets are given.

Experimental setup

Builds on: linear dynamic SCM single-door criterion d-separation

Closest papers by meaning

• Identifying Macro Causal Effects in C-DMGs over DMGs (NeurIPS 2025 · similarity 0.90)
• Identifying Macro Conditional Independencies and Macro Total Effects in Summary Causal Graphs with Latent Confounding (AAAI 2025 · similarity 0.88)
• Identification for Tree-Shaped Structural Causal Models in Polynomial Time (AAAI 2024 · similarity 0.88)
• Identifying Causal Effects Under Functional Dependencies (NeurIPS 2024 · similarity 0.87)
• CausalDynamics: A large‐scale benchmark for structural discovery of dynamical causal models (NeurIPS 2025 · similarity 0.85)
• Distinguishing Cause from Effect with Causal Velocity Models (ICML 2025 · similarity 0.85)